Discrepencies between free and top in reported memory usage

Hello all,

I've been google searching for several hours now and haven't found an answer so I decided it was time to post.

This is an SMP 2CPU Xeon machine running RedHat AS3.0-u02. The machine is running Oracle 9i, a java app server and several python servers.

I'm seeing different reports of memory usage from top and free.

free -m reports:

total used free shared buffers cached Mem: 4509 4482 26 0 134 2545 --+ buffers-cache: 1802 2706 Swap: 8189 805 7383

This would indicate about 2.5 GB of available (cached) memory which is in-line with my expectations.

However, when I look at the output of top, I see

PID USER PRI NI SIZE SWAP RSS SHARE D STAT %CPU %MEM TIME CPU COMMAND 31680 oracle 15 0 796M 4604 792M 790M 0M S 0.0 17.5 33:04 0 oracle 18311 app 21 0 826M 155M 671M 2752 0M S 0.0 14.8 39:36 0 java 2041 oracle 15 0 626M 2620 623M 623M 0M S 0.0 13.8 1:35 0 oracle 2899 oracle 16 0 580M 1324 579M 578M 0M S 0.0 12.8 24:25 3 oracle 14283 oracle 15 0 562M 3060 559M 559M 0M S 0.0 12.4 18:31 1 oracle 14281 oracle 15 0 546M 1472 545M 544M 0M S 0.0 12.0 18:51 0 oracle 14285 oracle 15 0 531M 2788 528M 527M 0M S 0.0 11.7 17:11 0 oracle 14287 oracle 15 0 520M 3840 516M 515M 0M S 0.0 11.4 17:16 1 oracle 3168 plus 15 oracle 15 0 387M 1152 386M 386M 95K S 0.0 8.5 9:41 1 oracle

Something is fishy right away as the memory usage is adding up to over 100%. I'm wondering if maybe this is adding in shared memory that oracle is using when performing it's calculations.

Could someone clarify why the descrepency between the two? How is it that free (and vmstat) sees 2.5 GB of cache but top (and ps for that matter) see Oracle using almost all the free memory on the system?

Thanks, -erik